SYMPLECTIC SINGULARITIES AND GEOMETRY OF GAUGE FIELDS
BANACH CENTER PUBLICATIONS, VOLUME 39
INSTITUTE OF MATHEMATICS
POLISH ACADEMY OF SCIENCES
WARSZAWA 1997
THE GROUP OF LARGE DIFFEOMORPHISMS
IN GENERAL RELATIVITY
DOMENICO GIULINI
Fakultät für Physik der Universität Freiburg,
Hermann Herder Straße 3, D-79104 Freiburg, Germany
E-mail: [email protected]
Abstract. We investigate the mapping class groups of diffeomorphisms fixing a frame at a
point for general classes of 3-manifolds. These groups form the equivalent to the groups of large
gauge transformations in Yang-Mills theories. They are also isomorphic to the fundamental
groups of the spaces of 3-metrics modulo diffeomorphisms, which are the analogues in General
Relativity to gauge-orbit spaces in gauge theories.
1. Introduction. The dynamics of General Relativity can be cast into the form of a
constraint Hamiltonian system. For this, the space-time M is assumed to be a topological
product Σ × R. Restricting to pure gravity, the unconstrained phase space is given by
the cotangent bundle, T ∗ (R), over the space R = Riem(Σ) of Riemannian metrics on
a 3-manifold Σ. Initial data thus consist of a 3-metric, gab , and a contravariant tensor
density of weight one, π ab . Once the space-time M has evolved from Σ, π ab is related to
√
the extrinsic curvature K ab of Σ in M via π ab = g(K ab − g ab Kcc ), where g = det{gab }.
But the data cannot be prescribed arbitrarily: they have to satisfy the two constraints
(1.1)
(1.2)
Da := −2∇b π ab = 0,
√
H := Gabcd π ab π cd − gR,
where ∇ denotes the Levi-Civita connection for gab and R its Ricci-scalar. Gabcd is a
symmetric, non-degenerate bilinear form of signature type (−, +, +, +, +, +), called the
DeWitt metric. Explicitly it reads:
1
(1.3)
Gabcd = √ (gac gbd + gad gbc − gab gcd ) .
2 g
1991 Mathematics Subject Classification: Primary 83C99; Secondary 57R50, 55S37.
This research was supported in parts by the Isaac Newton Institute, Cambridge (England),
and the Center of Geometry and Physics at the Pennsylvania State University. I thank these
institutions and the Banach Center in Warsaw for their hospitality.
The paper is in final form and no version of it will be published elsewhere.
[303]
304
D. GIULINI
Upon integration over Σ it defines a (weakly) non-degenerate form on T ∗ (R). Its metric
properties on the subspace of cotangent vectors satisfying (1.1) are quite complicated.
In particular, it is neither positive definite nor even of definite signature [8]. To set up
initial data we now have to
1. pick a 3-manifold Σ
2. find a pair (gab , π ab ) satisfying (1.1–2).
It is known that (1.1–2) have solutions on all topologies Σ [22]. In principle we therefore
have to consider all 3-manifolds.
This setting is similar to the Hamiltonian formulation of Yang-Mills theories: The
space of gauge potentials A there corresponds to the space of metrics R here, and the
group of gauge transformations (vertical bundle automorphisms) G there corresponds
to the diffeomorphism group D here. The constraint (1.1), called the diffeomorphisms
constraint, is analogous to the Gauss constraint in Yang-Mills theories. It should be read
as a momentum mapping for the action of the diffeomorphism group on phase space
which can be thought of as the lift of the action on configuration space:
D × R → R,
(1.4)
(φ, gab ) 7→ φ∗ gab .
d
A curve t → φt in D, where φ0 = id, defines a vector field ξ on Σ by ξ(p) = dt
|t=0 φt (p).
The infinitesimal version of (1.4) is then given by
d (1.5)
φ∗ gab = ∇a ξb + ∇b ξa ,
dt t=0 t
where the right-hand side is just the Lie derivative expressed in terms of ∇. The momentum map, P , applied to ξ, is simply obtained by contracting the canonical 1-form (“pdq”)
with the right-hand side of (1.5):
Z
Z
Z
3
ab
3
a
(1.6)
P (ξ) = 2
d x π ∇b ξa =
d x D ξa + 2
d2 x ξa π ab nb ,
Σ
Σ
∂Σ
where d3 x and d2 x denote coordinate volume elements and na the outward pointing
normal of ∂Σ. We thus see that the diffeomorphism constraint generates those diffeomorphisms for which the surface integral in (1.6) vanishes. For example, if Σ has only a single
connected boundary in an asymptotically flat region and if ξ asymptotically tends to a
translation or rotation, then the surface integral just corresponds to the total momentum
or angular momentum respectively. The corresponding motions are not generated by the
constraints. Only the asymptotically trivial ones are.
From now on we restrict Σ to have one regular end, i.e., there exists a compact set
K ⊂ Σ whose complement is homeomorphic to the complement of a 3-ball in R3 . This
idealizes the case of an isolated gravitational system. The regularity condition allows to
1-point compactify Σ to a compact manifold Σ = Σ ∪ ∞, where the added point is called
∞. Although physically we are interested in Σ it is easier, and for our purposes admissible,
to carry out all constructions on Σ. We thus take R to be the space of 3-metrics on Σ,
and D(Σ) its diffeomorphism group. Unless stated otherwise, we understand this group
to consist of orientation preserving diffeomorphisms only. R(Σ) is topologically a positive
open cone in a vector space and hence contractible. Since the diffeomorphism group ‘looks’
differently for different Σ it is given the argument to indicate this dependence. There are
GROUP OF LARGE DIFFEOMORPHISMS
305
two more diffeomorphism groups that we will be interested in:
(1.7)
D∞ (Σ) = {φ ∈ D(Σ) / φ(∞) = ∞},
(1.8)
DF (Σ) = {φ ∈ D∞ (Σ) / φ∗ |∞ = id}.
Topologically we do not loose anything if we replace D(Σ) with D∞ (Σ) and the asymptotically trivial diffeomorphisms of Σ with DF (Σ). These spaces are pairwise homotopy
equivalent. The diffeomorphism constraint generates the identity component of asymptotically trivial diffeomorphisms, i.e. the identity component DF0 (Σ) of DF (Σ). (Generally we
indicate identity components by an upper case 0.) The Hamiltonian formalism requires
to regard two points in the same orbit of the group generated by the diffeomorphism
constraint as physically identical. But there is still an action of the quotient group,
(1.9)
S(Σ) = DF (Σ)/DF0 (Σ).
Strictly speaking it is not required by the Hamiltonian formalism to also identify S(Σ)equivalent points. This is completely analogous to the situation in Yang-Mills theories,
where ‘large’ gauge transformations are not generated by the Gauss constraint. This
might lead to interesting physical consequences in the quantum theory. An example is
discussed in [5]. In the classical theory it would be appropriate to quotient by the S(Σ)action if we agreed that the functions on phase space which we want to use as observables
cannot separate any two points in one orbit of S(Σ). This is in fact usually assumed, so
that the classical configuration space is then given by Q(Σ) = R(Σ)/DF (Σ), which is in
fact the base of a DF (Σ)-principal-fibre-bundle [3]:
DF (Σ)
(1.10)
−→
R(Σ)


y
Q(Σ)
The contractibility of R(Σ) in the associated exact sequence for homotopy groups directly
leads to
(1.11)
π1 Q(Σ) ∼
= DF (Σ)/DF0 (Σ) = S(Σ),
(1.12)
πn Q(Σ) ∼
= πn−1 DF (Σ) , ∀n > 1.
Some calculations and results of higher homotopy groups (1.12) may be found in [6]. In
this paper we are concerned with the groups in (1.11).
If we did not quotient by S(Σ) we would have obtained as true configuration space
the universal cover Q(Σ) of Q(Σ). A classically equivalent procedure is to work with
Q(Σ) and adding the requirement that all observables must commute with the action of
S(Σ). This formulation allows us to stick with the simply connected configuration space
Q(Σ). In finite dimensions we would transfer this to the quantum theory as follows [7]:
take as Hilbert space H the L2 -space on Q(Σ) with some S(Σ)-invariant measure. Take
as algebra of observables the commutant of S(Σ). This induces a superselection structure
with sectors classified by the inequivalent irreducible unitary representations of S(Σ). For
this reasons, and for the obvious reasons in the classical theory, we are interested in the
group S(Σ) and its representations.
306
D. GIULINI
Obviously, the quantum mechanical motivation as outlined above has at least two
serious problems. The first is that the infinite dimensional space Q(Σ) is almost certainly
not an appropriate measure space and therefore cannot be used to construct H. It would
have to be replaced by something else which densely contains Q(Σ) as proper subset.
This goal is pursued in the works of Ashtekar and collaborators [1]. But what persists
to happen is that S(Σ) acts on the Hilbert space and is required to commute with
all observables. Our motivation is therefore still valid. The second difficulty is that the
discrete group S(Σ) is generically infinite and non-Abelian so that we cannot count on a
general representation theory. Note that this is a major difference to the situations usually
encountered in Yang-Mills theories. There the group of large gauge transformations is
typically Abelian and the representation theory is simple. For example, the famous θangle just labels the irreducible unitary representations of Z. Some equivalents of the
θ-angle in quantum-gravity have been studied in the literature [11][10]. In the sequel of
this paper we want to report on some approaches and results to understand the general
structure of the group S(Σ).
2. 3-manifolds. We consider a general closed, connected, and oriented 3-manifold
Σ. Up to permutation of factors it can be uniquely written as a finite connected sum
of prime 3-manifolds, simply called primes (see [13] for more information on the general
material that follows; we denote the connected sum by ]):
(2.1)
Σ=
n+l
]
i=1
Πi =
n
]
i=1
l
]
Pi ]
S2 × S1 ,
i=1
where amongst the general primes, Πi , we have notationally separated the irreducible
ones, Pi , from the ‘handles’, S 2 × S 1 . Recall that a 3-manifold is called irreducible if
every embedded 2-sphere bounds a 3-ball. Irreducibility implies a trivial second homotopy
group. The handle is the only reducible prime and connected sums are of course always
reducible. The known primes with finite fundamental group are all of the form S 3 /G
where G is a finite subgroup of SO(4) with proper action on S 3 . The irreducible ones
with infinite fundamental group are K(π, 1) spaces, i.e., their higher-than-first homotopy
groups are all trivial. Amongst those is the much better studied subclass of sufficiently
large K(π, 1)’s, i.e. those which contain incompressible surfaces. Simple examples are
those of the form S 1 × Rg , where Rg is a genus g surface, or the six orientable flat 3manifolds. It is conjectured but unproven that all primes with finite fundamental group
are covered by S 3 and all the irreducible ones with infinite fundamental group by R3 . If
the Poincaré conjecture were true the former (but not the latter) would be implied. For
convenience we shall in the sequel eliminate the uncertainty caused by the open Poincaré
conjecture and restrict to primes not covered by homotopy spheres and their connected
sums only. In (2.1) each Pi is connected to the rest of Σ along a 2-sphere, Si , 1 ≤ i ≤ n,
whereas each handle is connected along two 2-spheres, Sij , 1 ≤ i ≤ l, j = 1, 2, which
should be thought of as the boundaries of the cylinder [0, 1] × S 2 [16][9].
The fundamental group of the connected sum is given by the free product of the
individual ones:
GROUP OF LARGE DIFFEOMORPHISMS
(2.2)
π1 (Σ, ∞) =
n
∗
i=1
π1 (Pi ) ∗
l
∗
307
Z .
i=1
It is useful to visualize Σ in terms of the connected sum (2.1). Roughly speaking,
Σ is divided into the regions interior to the primes (called interior regions) and their
complement (called exterior region). The interior regions are connected to the exterior
region along the pairwise disjoint 2-spheres Si and Sij . This is explained in more detail
in [16][9]. This allows to divide the diffeomorphisms into three different types:
1. those with support in the interior region (called interior diffeomorphisms),
2. those which leave the exterior and interior region setwise invariant (called exterior
diffeomorphisms), like the exchange of two diffeomorphic primes,
3. special diffeomorphisms not in 1. and 2., called exterior slides.
These will be explained below.
An important tool in analysing S(Σ) is the following homomorphism
(2.3)
0
h∞ (Σ) : D∞ (Σ)/D∞
(Σ) → Aut(π1 (Σ, ∞)),
which one obtains by composing the loop γ that represents the element in the homotopy
group with φ ∈ D∞ (Σ) to φ ◦ γ. (This also works for higher homotopy groups.) Useful
information on S(Σ) is then obtained by investigating the kernel and image of h∞ (Σ).
For later notational convenience we also introduce the map hF which is defined just like
0
h∞ but with DF /DF0 replacing D∞ /D∞
.
Let us describe some of the important diffeomorphisms. First we turn to the internal
diffeomorphisms. Take an inner collar neighbourhood of Si , and define a diffeomorphisms
by a relative 2π-rotation of its boundary 2-spheres, and extend it to the rest of the
manifold by the identity. This is called a rotation of Pi . If this diffeomorphism is not
isotopic to the identity within the class of inner diffeomorphisms of Pi holding Si fixed,
we call Pi spinorial. Amongst the known primes all but the lens spaces and the handle
are spinorial and a connected sum is spinorial iff at least one prime is [9]. For spinorial
0
manifolds S(Σ) is a central Z2 -extension of D∞ (Σ)/D∞
(Σ). In the non-spinorial case
they are isomorphic [6]. The same relation hold for the kernels of hF (Σ) and h∞ (Σ).
Similarly, for a handle we can define an internal diffeomorphism by a relative 2π-rotation
of Si1 and Si2 , extended by the identity. This is called a twist of the i’th handle. It is
not isotopic to the identity and generates S(S 2 × S 1 ). Rotations of spinorial primes and
twists each generate a Z2 subgroup of S(Σ). Moreover, since they rotate parallel to 2spheres, they are in the kernel of h∞ (Σ). Clearly any two internal diffeomorphisms of
different primes commute. Hence there is a homomorphism of the direct product of all
S(Πi ) into S(Σ). Hendriks and McCullough have proven that this homomorphism is in
fact an injection [15]:
(2.4)
I:
n
Y
S(Pi ) × Zl2 ,→ S(Σ).
i=1
Applying a more general result of McCullough’s [17] to the present situation shows that
308
D. GIULINI
the kernel of hF (Σ) lies in the image of the map I, that is
(2.5)
ker((hF (Σ)) =
l
Y
i=1
n
Y
Z2 ×
Ki ,
i=1
where Ki denotes the kernel of hF in S(Pi ), which, we recall, is a Z2 -extensions of the
0
kernel of h∞ (Pi ) in D∞ (Pi )/D∞
(Pi ) if Pi is spinorial, and isomorphic otherwise. On the
other hand, suppose two diffeomorphisms d, d0 ∈ D∞ (Pi ) have the same image under
h∞ (Pi ). Irreducible primes have π2 = 0 and those of the form S 3 /G have π3 = Z whereas
K(π, 1) primes have π3 = 0. This implies that d and d0 are homotopic (see e.g. Lemma
5.1 in [21]). But for Pi for which homotopy implies isotopy – we call it the HI-property
and manifolds with this property HI-manifolds – this implies the triviality of Ki for nonspinorial Pi and Ki ∼
= Z2 if Pi is spinorial. No non-HI-prime is known so far (though
non-HI connected sums! [4]). But some cases are undecided and we refer to Theorem A1
in [6] for a list of which primes have so far been proven to be of HI type. For them we
have the
Proposition 1. Let Σ be composed of HI-primes of which k are spinorial. Then
Qk+l
ker(hF (Σ)) = i=1 Z2 , generated by rotations of spinorial primes and twists of handles.
The external diffeomorphism which we have to mention are the exchanges of diffeomorphic primes and the spins of handles. The latter just correspond to a mutual exchange
of the two handle-ends. More details are given in [9].
Next we describe slide diffeomorphisms: Given a loop γ : [0, 1] 3 s → Σ with toroidal
tubular neighbourhood T ⊂ Σ coordinatized by (r, θ, ϕ) ∈ [0, 2] × [0, 2π) × [0, 2π), where
r is a radial coordinate parametrizing coaxial 2-tori Tt = {r = t} with axis T0 = γ. θ
and ϕ coordinatize the toroidal latitudes and longitudes respectively. Let λ : R → [0, 1]
be a C ∞ step function: λ(x) = 0 for x ≤ 0, λ(x) = 1 for x ≥ 1. We define a 1-parameter
family of diffeomorphisms φt of Σ as follows:
p
for p ∈ Σ − T,
(2.6)
φt (p) =
(r, θ, ϕ + t2πλ(2 − r)) for p = (r, θ, ϕ) ∈ T .
The diffeomorphism φ1 = φt=1 is the identity inside T1 . It is called a slide along γ.
Suppose ∞ is on γ. Then φ1 ∈ DF (Σ) and the action of φ1 on π1 (Σ, ∞) is by conjugation
with [γ], the class of γ in π1 (Σ, ∞). This shows that
(2.7)
Inn(π1 (Σ, ∞)) ⊂ Im(h∞ (Σ)),
where we could have replaced h∞ with hF . Next suppose we slightly change the construction above in that one of the spheres Si (or Sij ) now sits inside T1 . We then call φ1
a slide of Pi (or the j’th end of handle i) along γ. This diffeomorphism is isotopic to the
identity if γ is contractible, but not otherwise [16].
An important theorem now states that S(Σ) is generated by internal, exchange, and
slide diffeomorphisms. The slides we need to consider are slides of primes or handle-ends
along non-contractible loops. Consequently, S(Σ) is finitely generated if each S(Pi ) is.
Lucid proofs of these important results may be found in [16].
The individual groups S(Pi ) have been calculated for many irreducible primes. See
[21] for the spherical primes S 3 /G. In the next section we show how to obtain a general
GROUP OF LARGE DIFFEOMORPHISMS
309
result for the class of sufficiently large K(π, 1) primes.
3. S(P ) for P sufficiently large K(π, 1). Let for the moment P be an irreducible
prime. The spaces D∞ (P ) and D(P ) are related by the fibration
i
D∞ (P ) −→
(3.1)
D(P
 )
p
y
P
p(φ) := φ(∞)
whose associated exact sequence for homotopy groups ends with (dropping the arguments
P for the moment)
(3.2)
p∗
∂
i
∗
· · · → π1 (D∞ ) → π1 (D) → π1 →∗ π0 (D∞ ) →
π0 (D) → 1.
The map ∂∗ is as follows: Take a class [γ] ∈ π1 (P, ∞) and a representing loop, γt , at ∞.
Take a lift, φt , of γt in D∞ starting at the identity map, φ0 = id, so that φt (∞) = γt . The
0
end point, φ1 , defines a class [φ1 ] ∈ π0 (D∞ ) := D∞ /D∞
, and one defines ∂∗ ([γ]) := [φ1 ].
This is indeed a well defined map since it is independent of the representative γt of [γ].
The lift φt may of course just be given by a slide along γ, as defined in (2.3). As noted
above, h∞ ([φ1 ]) just corresponds to conjugation by [γ], so that h∞ ◦ ∂∗ = Ad, where Ad
denotes the ‘adjoint-homomorphism’ from π1 to Inn(π1 ). Its kernel is Cπ , the centre of
π1 . Exactness of (3.2) implies that the image of p∗ lies within that kernel:
Im p∗ ⊂ Cπ .
(3.3)
This allows to write the inner automorphisms as a double quotient
. C
π
∼ π1
(3.4)
Inn(π1 ) =
=: q (π1 / Im p∗ ) ,
Im p∗ Im p∗
where q just denotes the quotient map with respect to Cπ / Im p∗ . Let h : D/D0 → Out(π1 )
be the non-based version of h∞ . Diagram (3.2) can now be completed by the maps q, h∞
and h with two commutative squares:
∂∗
i∗
π1
1 −→
−→
π0 (D∞ ) −→
π0 (D) −→ 1
Imp∗





(3.5)
yq
yh∞
yh
i0
p0
1 −→ Inn(π1 ) −→ Aut(π1 ) −→ Out(π1 ) −→ 1
Commutativity of the left square follows from the discussion above whereas commutativity of the right square is obvious. (3.3) implies that q is surjective which implies for the
diagram that h∞ is surjective if and only if h is. If we now restrict P to be an irreducible
HI-prime we can show
Lemma 1. Let P be an irreducible HI-prime, then q is an isomorphism.
P r o o f. For irreducible HI primes h∞ is injective. From (3.5) h∞ ◦ ∂∗ = i0 ◦ q with
injective maps h∞ , ∂∗ and i0 . So q must be injective. But q is a quotient map with respect
to Cπ / Im p∗ . So Im p∗ = Cπ and q is an isomorphism.
Hatcher has proven [12] that for sufficiently large K(π, 1) primes (which all have the
HI property) h is an isomorphism if D includes orientation reversing diffeomorphisms (if
existent). Generalizing for the moment to this case, diagram (3.5) then implies that h∞
310
D. GIULINI
is also an isomorphism. Taken together with the fact that K(π, 1) primes are spinorial
(this is e.g. shown in [14] or [20]) this implies
Proposition 2. Let P be a sufficiently large K(π, 1) prime. Then S(P ) is a central
Z2 extension of Aut+ (π1 (P )).
Here we had to account for the possibility that P allows for orientation reversing
diffeomorphisms all of which act non-trivially on the fundamental group. Since diffeomorphisms in DF (P ) are necessarily orientation preserving, we can in this case only
reach an index 2 subgroup of Aut(π1 (P )) which we denoted by an upper case + sign.
This difference does not exist if P does not allow for orientation reversing diffeomorphisms
(this is the generic case, see e.g. [9] for collective information on this point), or if there
exists an orientation reversing diffeomorphism acting trivially on the fundamental group,
like e.g. for the handle. For example, for the 3-torus T 3 we have Aut(Z3 ) = GL(3, Z),
Aut+ (Z3 ) = SL(3, Z) and one may show that S(T 3 ) is given by the Steinberg group
St(3, Z) (see [19] for more information on the Steinberg groups).
4. General Remarks. For a physicist it is tempting to think of the manifold Σ as
being composed of elementary objects, the primes, just like a collection of N = n + l particles from d different species each with its own characteristic internal symmetry group
Gr , 1 ≤ r ≤ d (in doing this we think of the spin of a handle as internal diffeomorphism).
In this analogy diffeomorphic primes correspond to particles of one species and the ‘internal’ symmetry groups Gr to the groups S(Πr ). Let there be nr primes in the r’th
Pd
diffeomorphism class so that r=1 nr = N . In the particle picture the total symmetry
group would be a semidirect product of the internal symmetry group GI with the external
symmetry group GE . These are given by:
(4.1)
GI =
n1
Y
i=1
(4.2)
G1 × · · · ×
nd
Y
Gd ,
i=1
GE =Sn1 × · · · × Snd .
There is a homomorphism θ : GE → Aut(GI ), defined by θ = θ1 × · · · × θd , where
nj
Y
θj : Snj → Aut
Gj ,
(4.3)
i=1
σ 7→ θj (σ) : (g1 , . . . , gnj ) 7→ (gσ(1) , . . . , gσ(nj ) ).
The semidirect product GI ×θ GE =: GP , which we call the particle group, is now defined
Q nj
via the multiplication law as follows: Let γj ∈ i=1
Gj , j = 1, ..., d, then
(4.4)
(γ10 , . . . , γd0 ; σ10 , . . . , σd0 ) (γ1 , . . . , γd ; σ1 , . . . , σd )
= (γ10 [θ1 (σ10 )(γ1 )], . . . , γd0 [θd (σd0 )(γd )] ; σ10 σ1 , . . . , σd0 σd ) .
From the discussion in Section 2 it is clear that the group GP is a subgroup of S(Σ)
generated by internal and external diffeomorphisms. However, we also had to consider
slides which were neither internal nor external and which are not compatible with the
particle picture just used. As regards our understanding of S(Σ), it seems that one of
GROUP OF LARGE DIFFEOMORPHISMS
311
the most interesting questions is to ask for the rôle of slides in building up S(Σ). For
example, we may ask for the normal subgroup NS generated by slides and whether it
overlaps with the particle group GP , i.e., whether NS ∩ GP 6= {1}. If so, the quotient
S(Σ)/NS will be a non-trivial factor group of GP . Otherwise S(Σ) will be a semidirect
product of NS with GP . In the sequel of this paper we attempt to answer this question.
5. Examples. In this section we wish to develop some feeling for the question just
posed by presenting two examples of manifolds each consisting of a N > 2 fold connected
sum of diffeomorphic primes. In the first example the primes will be handles:
(5.1)
Σ=
N
]
S2 × S1,
i=1
so that the fundamental group of Σ is just the free group on N generators g1 , . . . , gN :
N
(5.2)
π1 (Σ) =
∗Z=F
N.
i=1
We represent each gi by a loop based at ∞ which enters the i’th handle through Si1 and
leave it through Si2 (call it the positive direction). The group Aut(FN ) can be generated
by the four operations
(5.3a)
P : [g1 , g2 , g3 , . . . , gN ] 7→ [g2 , g1 , g3 , . . . , gN ],
(5.3b)
Q : [g1 , g2 , g3 , . . . , gN ] 7→ [g2 , g3 , g4 , . . . , gN , g1 ],
(5.3c)
O : [g1 , g2 , g3 , . . . , gN ] 7→ [g1−1 , g2 , g3 , . . . , gN ],
(5.3d)
U : [g1 , g2 , g3 , . . . , gN ] 7→ [g1 g2 , g2 , g3 , . . . , gN ],
whose interpretation is as follows: P exchanges handle no. 1 and handle no. 2, Q exchanges
all N handles in cyclic order, O spins handle no. 1, and U slides the second end of the first
handle through the second handle in a negative direction. This means that by inspection
we have just proven that in the present case the map h∞ (Σ) in (2.3) is surjective. From
QN
(2.5) we know that its kernel and hence that S(Σ) is an extension of Aut(FN ) by i=1 Z2 .
It is also easy to describe the extension explicitly. First of all, P and Q alone generate
the permutation group SN of N objects which is a sub- but no factor-group of Aut(FN ),
and P, Q, O generate a subgroup of order 2N N ! whose interpretation is that of a particle
group with internal Z2 symmetry (the spin). A presentation of the full group is obtained
by adding a generator T which generates a twist on the first handle. It commutes with
U and O, squares to the identity, and has the same relations with P, Q as O has. Since
we are given an explicit presentation of Aut(FN ) in terms of P, Q, O, U [2] we can now
give an explicit presentation of S(Σ) in terms of P, Q, O, T, U . We use the symbol V to
denote either O or T so that a relation containing V is meant to be valid with V = O
and V = T . A ↔ B says that A commutes with B:
(5.4a)
P2 = V 2 = E
(5.4b)
(QP )N −1 = QN
(5.4c)
P ↔ Q−i P Qi ,
for 2 ≤ i ≤ N/2
312
D. GIULINI
(5.4d)
V ↔ Q−1 P Q
(5.4e)
V ↔ QP
(5.4f)
(P V )4 = E
(5.4g)
T ↔O
(5.4h)
U ↔T
(5.4i)
U ↔ Q−2 P Q2
U ↔ QP Q
(5.4j)
−2
−1
for N > 3
PQ
OQ2
(5.4k)
U ↔Q
(5.4l)
U ↔ Q−2 U Q2
(5.4m)
U ↔ OU O
(5.4n)
U ↔ P Q−1 OU OQP
(5.4o)
U ↔ P Q−1 P QP U P Q−1 P QP
(5.4p)
(P Q−1 U Q)2 = U Q−1 U QU −1
(5.4q)
U −1 P U P OU OP O = E
for N > 3
Most of the ↔ relations have a trivial meaning. For example, (5.4j) just says that
U commutes with a cyclic permutation of gi ’s for i > 2. But this is obvious since their
supports may be chosen to be disjoint. The most remarkable relation is (5.4q) which
can be rewritten in the form OP = U −1 (P U P −1 )(OU O−1 ). Hence OP is in the normal
subgroup generated by slides. Adding the relation U = E will therefore not lead to the
particle group. In fact we have
Theorem 1. The quotient with respect to the normal subgroup generated by slides is
Z2 × Z2 generated by T and P .
P r o o f. Setting U = E in (5.4q) leads via (5.4a) to P = O. (5.4e) for V = O then
implies P ↔ Q and (5.4b) implies Q = P N −1 , i.e., Q = E for N odd and Q = P for N
even. We are thus left with two involutive commuting generators T and P .
By simply checking the relations one may moreover prove [9]
Theorem 2. Given a representation ρ of S(Σ), the following conditions are equivalent:
1.
2.
3.
4.
ρ is Abelian,
slides are represented trivially,
ρ correlates P and O, i.e. ρ(P ) = ρ(O),
slides and exchanges commute under ρ.
As regards this theorem we mention that the case N = 2 behaves rather differently [9].
In our second example we take as prime the real projective space RP 3 . The setting
is very similar to the previous case so that we can be brief. The fundamental group is
the free product on N involutive generators g1 , . . . , gN , gj2 = E. There is no non-trivial
internal diffeomorphism (e.g. [21]), i.e. S(P R3 ) = E so that the kernel of h∞ (Σ) trivial.
S(Σ) is generated by the operations:
(5.5a)
P : [g1 , g2 , g3 , . . . , gN ] 7→ [g2 , g1 , g3 , . . . , gN ],
313
GROUP OF LARGE DIFFEOMORPHISMS
(5.5b)
Q : [g1 , g2 , g3 , . . . , gN ] 7→ [g2 , g3 , g4 , . . . , gN , g1 ],
(5.5c)
U : [g1 , g2 , g3 , . . . , gN ] 7→ [g2−1 g1 g2 , g2 , g3 , . . . , gN ],
and the relations are given by
(5.6a)
P2 = U2 = E
(5.6b)
(QP )N −1 = QN = E
(5.6c)
P ↔ O−i P Qi ,
(5.6d)
(5.6e)
−2
U ↔Q
2
PQ ,
U ↔ QP Q
−2
−1
for , 2 ≤ i ≤ N/2
for , N > 3
PQ
U Q2
(5.6f)
U ↔Q
(5.6g)
U ↔ Q−1 P U P Q
(5.6h)
Q−1 U QU Q−1 U Q = P Q−1 U QP U P Q−1 U QP,
for N ≥ 3.
Of course, the first three relations again imply that P, Q generate the permutation group
SN which here is identical to the particle group GP . All ↔ relations can be understood
from simple arguments saying that the diffeomorphisms representing the left and right
sides can be chosen to have disjoint support. Relation (5.6h) is less obvious. If µij represents the unique slide of prime j through prime i, then it can be expressed in the
form: µ31 µ21 µ31 = µ32 µ21 µ32 . (Relation (5.4p) has a similar interpretation using slides of
handle-ends.) In particular, it is a relation purely amongst slides. Setting U = E in (5.6h)
leads to no additional relations for P and Q, so that now we indeed have S(Σ)/NS = GP .
How can such a different behaviour be explained? Formally, a major difference between
handles and irreducible primes is that for handles we have the possibility to slide only
their ends rather than the whole prime. This results is the difference of (5.3d) and (5.5c):
a conjugation on generators, like in (5.5c), can be generated from (5.3) (sliding both ends)
but a mere left or right multiplication on generators, like in (5.3d), cannot be generated
from (5.5). This suggests that the presence of handles is responsible for the particle group
not being a quotient. In the next section we report some work that proves this conjecture.
6. The General Case. We recall that in the general case π1 (Σ) is a free product of
the groups π1 (Pi ) and some Z’s. S(Σ) is to be built from ker(h∞ (Σ)) and Im(h∞ (Σ)) ⊆
Aut(π1 (Σ)). The reason why we can make statements about the general case lies in the
happy fact that we are explicitly given a presentation for the automorphism group of free
products. We use the form given in [18] (called the Fuks-Rabinovitch presentation), in
which the generators of Aut(π1 (Σ)) are given by those of Aut(π1 (Pi )) and others which
correspond to spin-, exchange-, and slide-diffeomorphisms. The lack of surjectivity of
h∞ (Σ) on Aut(π1 (Σ)) resides entirely in the lack of surjectivity of its restrictions h∞ (Πi ) :
S(Πi ) → Aut(π1 (Πi )). For example, using Proposition 2, we may infer surjectivity if Σ
is a connected sum of handles and sufficiently large K(π, 1) primes. Since these are all
HI-primes, we can combine this with the first example of the previous section and infer
Theorem 3. Let Σ be a connected sum of l handles and n sufficiently large K(π, 1)
Qn+l
primes. Then S(Σ) is an extension of Aut+ (π1 (Σ)) by i=1 Z2 .
314
D. GIULINI
Like in the N -handle example the extension would not be difficult to describe. The Z2 ’s
correspond to twists and rotations which commute with other internal diffeomorphisms
and slides, but are acted upon by exchanges in the obvious way. Together with the explicit
presentation of Aut(π1 (Σ)) this is then sufficient information to write down a presentation
of S(Σ).
We now address the question posed at the end of the previous section. To do this one
has to explicitly go through the Fuks-Rabinovitch relations as given in [18]. We will not
repeat them here but rather refer the reader to the article by McCullough and Miller.
We shall adopt their numbering of the relations. Amongst these relations are four of
particular interest for us whose implications we now wish to explain:
Relation 30: Given any connected sum containing (amongst possible other primes)
two handles Hi , Hj and another prime Πk (which may be a handle or not), then slides of
a handle-end of Hj through Πk can be written as a commutator of slides.
Relation 47: Given any connected sum containing (amongst possible other primes) a
prime Πi (which may be a handle or not), a handle Hj , and an irreducible prime Pk ,
then any slide of Pk through Πi can be written as a commutator of slides.
These two observations imply
Theorem 4. If Σ contains at least three handles then slides generate a perfect subgroup NS of S(Σ).
This generalizes our observation concerning slides made on the first example above.
Relation 31: Given any connected sum containing (amongst possible other primes)
two handles Hi and Hj , then a spin of Hj followed by an exchange of Hi and Hj is
generated by slides of handle-ends of Hi through Hj and of Hj through Hi .
Relation 48: Given any connected sum containing (amongst possible other primes) an
irreducible prime Pi and a handle Hj , then internal slides of Pi are generated by external
slides of Pi through Hj and ends of Hj through Pi .
General Observation: The last two relations are the only ones that imply additional
relations for non-slide generators when the slides generators are set equal to the identity
E. This implies:
∼
Theorem 5. If Σ contains no handle in its prime decomposition, then S(Σ))/NS =
GP . In fact, S(Σ) is a semi-direct product of NS with GP .
Note that this implies in particular that all (irreducible) representations of GP extend
to (irreducible) representations of S(Σ). We can now understand our two examples in the
previous section as extreme cases for the last two theorems.
References
[1]
[2]
A. A s h t e k a r, J. L e w a n d o w s k i, D. M a r o l f, J. M o u r ã o, and T. T h i e m a n n, Quantization of diffeomorphism invariant theories with local degrees of freedom, J. Math. Phys.
36 (1995), 6456–6493; gr-qc/9504018.
H. S. M. C o x e t e r and W. O. J. M o s e r, Generators and relations for discrete groups
(second edition), Springer Verlag, Berlin–Göttingen–Heidelberg–New York, 1965.
GROUP OF LARGE DIFFEOMORPHISMS
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
315
A. E. F i s c h e r, Resolving the singularities in the space of Riemannian geometries,
J. Math. Phys. 27 (1986), 718–738.
J. F r i e d m a n and D. W i t t, Homotopy is not isotopy for homeomorphisms of 3-manifolds,
Topology 25 (1986), 35–44.
D. G i u l i n i, Asymptotic symmetry groups of long-ranged gauge configurations, Modern
Phys. Lett. A 10 (1995), 2059–2070.
D. G i u l i n i, On the configuration space topology in general relativity, Helv. Phys. Acta 68
(1995), 87–111.
D. G i u l i n i, Quantum mechanics on spaces with finite fundamental group, Helv. Phys.
Acta 68 (1995), 438–469.
D. G i u l i n i, What is the geometry of superspace? , Phys. Rev. D (3) 51 (1995), 5630–5635.
D. G i u l i n i, 3-Manifolds for relativists, Internat. J. Theoret. Phys. 33 (1994), 913–930.
D. G i u l i n i and J. L o u k o, Diffeomorphism invariant states in Wittens 2 + 1 quantum
gravity on R × T 2 , Classical Quantum Gravity 12 (1995), 2735–2745.
D. G i u l i n i and J. L o u k o, Theta-sectors in spatially flat quantum cosmology, Phys. Rev.
D (3) 46 (1992), 4355–4364.
A. H a t c h e r, Homeomorphisms of sufficiently large P 2 -irreducible 3-manifolds, Topology
15 (1976), 343–347.
J. H e m p e l, 3-Manifolds, Ann. of Math. Stud. 86, Princeton University Press, 1976.
H. H e n d r i k s, Application de la théorie d’obstruction en dimension 3, Bull. Soc. Math.
France Mém. 53 (1977), 81–195.
H. H e n d r i k s and D. M c C u l l o u g h, On the diffeomorphism group of a reducible 3manifold , Topology Appl. 26 (1987), 25–31.
D. M c C u l l o u g h, Mappings of reducible manifolds, in: Geometric and Algebraic Topology, Banach Center Publ. 18 (1986), 61–76.
D. M c C u l l o u g h, Topological and algebraic automorphisms of 3-manifolds, in: Groups of
Homotopy, Equivalences and Related Topics, R. Piccinini (ed.), Lecture Notes in Math.
1425, Springer, Berlin, 1990, 102–113.
D. M c C u l l o u g h and A. M i l l e r, Homeomorphisms of 3-manifolds with compressible
boundary, Mem. Amer. Math. Soc. 344 (1986).
J. M i l n o r, Introduction to algebraic K-theory, Ann. of Math. Stud. 72, Princeton University Press, 1971.
S. P. P l o t n i c k, Equivariant intersection forms, knots in S 4 , and rotations in 2-spheres,
Trans. Amer. Math. Soc. 296 (1986), 543–575.
D. W i t t, Symmetry groups of state vectors in canonical quantum gravity, J. Math. Phys.
27 (1986), 573–592.
D. W i t t, Vacuum space-times that admit no maximal slice, Phys. Rev. Lett. 57 (1986),
1386–1389.